École Nationale Supérieure de
Techniques Avancées

Résumé

Virtual reality applications including visual and gestural man-machine interaction requires the use of particle-based deformable and articulated models. However, they lack the relevant model building methods, in order to ensure realistic behaviours based on observations of real-world objects. To this end, our research aims at looking for reliable physical model building mechanisms, which use experimental kinematic data of real objects as an input. Building a physical-based model of an object thus divides into two main steps: • The first step is the capture of kinematic data. It may consist of extracting characteristic points trajectories from the images of a real scene, involving image sequence analysis techniques, or other ad-hoc experimental means [6]. • the second step consists of using these kinematic data into a specific physical model implementation scheme. This paper focuses on the second step. We propose an original method to automatically identify physical models built through using local masses and generalised springs.